Task-State EEG Signal Classification for Spatial Cognitive Evaluation Based on Multiscale High-Density Convolutional...

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Spatial cognitive ability assessment is crucial for understanding brain function.
• Electroencephalogram (EEG) signals contain valuable information about cognitive states.
• Developing accurate methods to evaluate cognitive training effects is essential.

Purpose of the Study:

• To propose a novel multi-scale high-density convolutional neural network (MHCNN) for classifying EEG signals.
• To assess the effectiveness of spatial cognitive training by analyzing task-state EEG data.
• To establish MHCNN as a reliable biological indicator for cognitive training outcomes.

Main Methods:

• Extracted EEG frequency band features using multi-dimensional conditional mutual information.
• Transformed multi-frequency band coupling features into multi-spectral images.
• Employed a Densenet-improved multi-scale convolutional neural network with two-scale convolution kernels.
• Optimized stochastic gradient descent for evaluating training effects.

Main Results:

• The proposed MHCNN achieved a highest accuracy of 98% in classifying EEG signals.
• MHCNN outperformed classical Convolutional Neural Network (CNN) and multi-scale CNN.
• The Theta-Beta2-Gamma frequency band combination demonstrated the best classification performance.
• Six frequency band combinations showed significant classification accuracy.

Conclusions:

• The MHCNN classification method is effective for assessing spatial cognitive training effects.
• MHCNN can serve as a valuable biological indicator for cognitive training.
• The proposed method has potential for broader applications in brain function evaluation.